Case Study
How Tilt moves at internet speed with ContextFlo
Tilt’s marketplace team turned week-long investigations into real-time answers without adding headcount.
About Tilt
Tilt is a livestream shopping platform where sellers auction products live to buyers. With thousands of sellers, thousands of daily transactions, and real-time decisions driving GMV, speed of insight is competitive advantage.
Data stack
Amplitude, Snowflake, dbt, Dagster
How ContextFlo Helped
- •Ops, PM, and finance leaders investigate issues end-to-end without analyst queues.
- •ContextFlo blends product, marketing, and payments data so answers live in one session.
- •Alert-driven workflows catch performance swings and trigger SQL automation instantly.
The Challenge
Fast-moving business meets slow data access.
Tilt operates at internet speed:
- •Sellers go live multiple times per day
- •Conversion rates shift by the hour
- •Payment issues need immediate investigation
- •Fraud patterns emerge in real-time
But their data team couldn’t keep up:
- •3-5 people could write production SQL
- •Everyone else waited days for analysis
- •Simple questions became backlog items
- •Critical decisions delayed while waiting for data
The breaking point:
When business metrics shift unexpectedly (as they do frequently in livestream commerce), the team needs to investigate across dozens of dimensions (media source, device type, seller segments, time periods) to find root cause. Traditional approaches would take weeks. They need answers in minutes.
The Solution
Self-service analytics powered by ContextFlo.
Tilt connected ContextFlo to their Snowflake warehouse and gitlab repos. Within Snowflake, they have both business data and data from external tools (eg Linear, Stripe, fb ads, tiktok ads). Within days, team members across product, operations, and leadership were running complex analyses independently.
Key capabilities they use:
- •Funnel analysis (install → signup → purchase)
- •Seller performance tracking (individual and segment-level)
- •Fraud detection (referral abuse, coupon exploitation)
- •Supply/demand matching (marketplace liquidity)
- •Payment reconciliation
- •Real-time anomaly investigation
Usage in Practice
Sample queries the team runs every week.
"Determine if we have a break in our payments funnel"
"Match daily signups' brand interests to brands actually auctioned that day, create auctions-per-signup ratio for top 10 brands over 90 days"
"Show GMV minute-by-minute compared to average for this day of week over past 4 weeks"
"Determine if a user '***' is abusing referral coupons"
Results
Usage metrics (last 30 days) and real business impact.
Volume:
- •~4k sql queries executed
- •2k unique analytical sessions
- •~100 queries per day average
Adoption in first two months:
- •27 active users (50% of team)
- •Used everyday in past month
- •Peak: 357 queries in a single day
Speed:
Questions that previously took 2-5 days now get answered in 2-5 minutes.
Real Business Impact
Rapid Investigation Example:
When key business metrics shift unexpectedly, Tilt uses ContextFlo to:
- 1.Identify the change within hours (not days)
- 2.Segment by media source, device, seller tier, time period
- 3.Test multiple hypotheses in parallel
- 4.Trace root causes across complex data relationships
- 5.Implement fixes based on data-driven insights
- 6.Use metric-trees to quickly see why there was a change in a top level metric like GMV by checking its dependencies.
Traditional approach: Days or weeks of data team backlog, decision paralysis, delayed response.
Fraud Detection:
Team member noticed unusual referral patterns, ran 16 queries investigating a specific user, identified systematic coupon abuse, prevented further losses within same day.
Seller Operations:
Operations team runs daily performance checks on various sellers, identifying:
- •Tier transitions
- •Schedule optimization opportunities
- •At-risk high-performers
- •New seller conversion patterns
All self-service, no data team bottleneck.
Marketplace Liquidity:
Marketplace team runs supply/demand matching analyses to ensure auctions stay competitive, balancing new seller onboarding with buyer demand patterns.
Why It Works
1. Context, Not Just Schema
ContextFlo auto-generates table context from their GitLab docs, understanding:
- •What "active user" means in Tilt's business
- •How orders relate to auctions relate to rooms
2. Agentic Workflow
For complex questions, ContextFlo:
- •Explores relevant tables
- •Reads business context
- •Iterates on queries
- •Validates results
- •Explains findings
The October investigation required 41 queries as ContextFlo systematically tested hypotheses. A human analyst would have done the same—but over weeks.
3. Trusted Results
Team members trust ContextFlo's analysis because:
- •Results are consistent and verifiable
- •Queries are transparent and reviewable
- •Actually faster than writing SQL manually
Key Metrics
| Active Users | 27 (50% of team) |
| Daily Usage | 30/30 days |
| Queries/Day | 125 average |
| Time to Insight | Minutes vs days |
| Data Team Reduction | 75% fewer routine requests |
What Tilt's Team Says
Hedi Ketari, Lead Data Scientist at Tilt:
"ContextFlo transformed how our team works with data. Instead of waiting days for answers, anyone can investigate complex questions independently. The speed and accuracy mean we can respond to business changes in real-time rather than after the fact."
The Bottom Line
Before ContextFlo
- ×3-5 people could analyze data
- ×2-5 day turnaround for simple questions
- ×Critical decisions delayed
- ×Data team = bottleneck
After ContextFlo
- ✓27 people analyze data independently
- ✓2-5 minute turnaround for complex analysis
- ✓Real-time decision making
- ✓Data team = strategic
The real value isn't just cost savings from increased productivity —it's competitive velocity. In livestream e-commerce, the company that identifies and fixes conversion drops fastest wins.
Ready to move at internet speed?
See how ContextFlo can transform your team's data access and decision-making velocity.